The platform hosts several descriptor extraction tools that enable multi-perspective modeling of protein-ligand interactions. Use this service to extract thousands of descriptors for key interactions.
Descriptors calculated by DDB or those dontated by other users can be downloaded for free. You can also download the programs used to generate such descriptors. Structures of protein-ligand complexes and their screening data can also be downloaded.
Perform automatic feature-subset-selection to filter descriptors from noise and irrelevant information. The filtering report produced using this service can give insights on the effect of different interactions for the given prediction task.
BUILD SCORING FUNCTIONS
Train state-of-the-art machine learning models using thousands of DDB descriptors to build accurate scoring functions. DDB’s fit/predict system will also generate detailed performance reports of ML scoring functions on independent protein-ligand complexes.
The platform provides three main types of descriptors:
Features that characterize the receptor primary and tertiary structure:
REPAST . RETEST .
Features that describe small molecules:
ECFP . PaDEL .
Features that characterize the protein-ligand complex as a whole:
AffiScore . AutoDock Vina . ChemGauss . Cyscore . DPOCKET . DSX . GOLD . LigScore . NNScore . RF-Score . SMINA . X-Score.
E.g. xscore, affiscore, chemgauss
thousand different descriptors
Docking, Screening, Scoring, Ranking
PDBbind, DUD, ChEMBL